Kubernetes pod disruption budgets during cluster-autoscaler scale-down

Running EKS with cluster-autoscaler and multiple node groups. We've set PDBs on our stateful workloads (minAvailable: 75%), but during scale-down events the autoscaler still evicts pods that violate the budget before new nodes are ready. The CA docs say it respects PDBs, but we're seeing a 30-60s window where pods get evicted before replacement nodes pass readiness. How are you configuring: - maxSurge / maxUnavailable on deployments with PDBs - CA's `scale-down-utilization-threshold` and `scale-down-delay-after-add` - topologySpreadConstraints to prevent single-AZ concentration during scale-down We're on K8s 1.29 with CA 1.28.x. Any war stories or configs that actually work in production?